Generative AI for architects
On October 5th 2023, I presented “Generative AI for Architects” to the Utrecht Architect Community Group in Amsterdam (slides here). The goal was to give fellow Salesforce architects a practical, non-hype grounding in generative AI: what large language models actually are, how Salesforce is building Einstein GPT on top of them, and where the wider ecosystem is heading.
What the talk covered
- A plain-language introduction to how large language models work, aimed at architects rather than data scientists
- How Einstein GPT fits into the Salesforce platform, and what’s different from “just calling an API”
- A live look at Einstein for Developers, the VS Code extension that brings generative AI into the Apex/LWC development workflow
- Where Salesforce’s models published on Hugging Face fit into the picture
- A walkthrough of GPT-powered features already shipping across the platform - see the Platform Demos walkthrough of GPT in Salesforce
Why architects specifically need this
Generative AI changes a few of the assumptions architects normally design around: data goes to a model (where, exactly?), outputs are non-deterministic (how do you test that?), and “build vs. buy” now includes “which foundation model, and via which trust layer?” None of that is exotic once you’ve seen it once - but it does need to be designed for deliberately, the same way we’d design for governor limits or sharing rules.
Go deeper
- Apex Hours: Einstein GPT for Developers
- Andrej Karpathy: Let’s build GPT: from scratch, in code, spelled out
- Databricks: Generative AI Fundamentals
- Salesforce: Einstein GPT release dates and the Winter ‘24 released features
Related reading
This talk built on Overview of Salesforce Einstein and Einstein GPT features, and on the broader landscape in Cloud AI development services & capabilities in 2023 and Salesforce Data Cloud.